Date of Award
5-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Chemistry and Chemical Engineering
First Advisor
Bo Wang
Second Advisor
James R. Brenner
Third Advisor
Nasri Nesnas
Fourth Advisor
Maria Pozo de Fernandez
Abstract
The field of metabolomics serves as a critical downstream "omics" approach,
providing a functional output of biological systems by examining the collection of small
molecules called metabolites. Despite its potential, the field faces significant challenges
in comprehensive metabolome measurement, metabolite identification, and complex
data interpretation. This dissertation addresses these limitations by advancing
metabolomics methodologies through the integration of machine learning for
environmental profiling, human health studies, and drug development applications.
The research encompasses several key objectives. First, it focuses on the
development of automated machine learning approaches, specifically artificial deep
learning neural network classification (ANNDL-DA), to improve two-dimensional (2D)
nuclear magnetic resonance (NMR) data interpretation and peak selection without
requiring extensive NMR expertise. Second, the study investigates the sex-specific and
organ-specific metabolic alterations induced by ethanol exposure in mice using NMR-
based metabolomics. Third, a metabolic profiling protocol was developed for botanical
sources often considered agricultural waste or invasive species, including bitter melon,
iv
palmer amaranth, and garlic scapes. These efforts serve to showcase to the public the
beneficial and medicinal properties of these weeds.
Methodologies employed across these studies include NMR spectroscopy, liquid
chromatography-mass spectrometry (LC-MS), and flash chromatography for high-
resolution separation. Key findings demonstrate that artificial deep learning neural
network classification (ANNDL-DA) provides superior accuracy in automated peak
selection compared to traditional methods. Moreover, ethanol exposure in adult mice
showed a different metabolic response in male and female mice as well variations on
the organ levels when fecal, liver, and serum samples were examined. Additionally,
botanical investigations identified a rich reservoir of beneficial flavonoids, supporting the
potential for repurposing invasive species into viable medical treatments for conditions
such as diabetes, high cholesterol, and cancer. Finally, co-authored studies highlighted
the significance of metabolic shifts in aging, Alzheimer's disease, and environmental
toxin exposure. Collectively, this work bridges regional knowledge gaps and provides
robust, automated tools to streamline future metabolomics research in environmental
and human health applications.
Recommended Citation
Pollak, Julie Pascal, "Novel Metabolomics Methodologies Development using Machine Learning Approaches and Applications in Environmental, Natural Compounds, and Human Health Studies." (2026). Theses and Dissertations. 1646.
https://repository.fit.edu/etd/1646
Comments
Copyright held by author